This is the mail archive of the
gcc@gcc.gnu.org
mailing list for the GCC project.
Re: [patch] Re: CSE improvement
- To: jfc at mit dot edu (John Carr)
- Subject: Re: [patch] Re: CSE improvement
- From: Joern Rennecke <amylaar at cygnus dot co dot uk>
- Date: Tue, 10 Mar 1998 13:34:38 +0000 (GMT)
- Cc: haible at ilog dot fr, egcs at cygnus dot com
>
> > Only on Sparc, not on i386. It depends on strict alignment of p1 and p2.
> > (If p1 and p2 differ by half a word, nothing can be said about *p1 after
> > the store.)
>
> C code which stores to partially overlapping locations is non-conforming.
What about complex? IIRC, it is supposed to work like an array of
float/double. Thus, if both p1 and p2 are pointers to a complex type and
point into a larger array of the base type, would it be legal for them to
overlap ?